Real-Time Imaging of Perivascular Transport of NanoparticlesDuring Convection-Enhanced Delivery in the Rat Cortex

CONOR P. FOLEY,1,4 NOZOMI NISHIMURA,2 KEITH B. NEEVES,3 CHRIS B. SCHAFFER,2

and WILLIAM L. OLBRICHT1,2

1School of Chemical and Biomolecular Engineering, Cornell University, Ithaca, NY 14853, USA; 2Department of BiomedicalEngineering, Cornell University, Ithaca, NY 14853, USA; 3Chemical Engineering Department, Colorado School of Mines,

Golden, CO 80401, USA; and 4Citigroup Biomedical Imaging Center, Radiology Department, Weill Cornell Medical College,516 E 72nd Street, New York, NY 10021, USA

(Received 2 July 2011; accepted 7 October 2011)

Associate Editor Daniel Elson oversaw the review of this article.

Abstract—Convection-enhanced delivery (CED) is a prom-ising technique for administering large therapeutics that donot readily cross the blood brain barrier to neural tissue. It isof vital importance to understand how large drug constructsmove through neural tissue during CED to optimize con-struct and delivery parameters so that drugs are concentratedin the targeted tissue, with minimal leakage outside thetargeted zone. Experiments have shown that liposomes, viralvectors, high molecular weight tracers, and nanoparticlesinfused into neural tissue localize in the perivascular spacesof blood vessels within the brain parenchyma. In this work,we used two-photon excited fluorescence microscopy tomonitor the real-time distribution of nanoparticles infused inthe cortex of live, anesthetized rats via CED. Fluorescentnanoparticles of 24 and 100 nm nominal diameters wereinfused into rat cortex through microfluidic probes. Wefound that perivascular spaces provide a high permeabilitypath for rapid convective transport of large nanoparticlesthrough tissue, and that the effects of perivascular spaces ontransport are more significant for larger particles thatundergo hindered transport through the extracellular matrix.This suggests that the vascular topology of the target tissuevolume must be considered when delivering large therapeuticconstructs via CED.

Keywords—Microfluidics, Two-photon microscopy, Neural

drug delivery.

INTRODUCTION

Convection-enhanced delivery (CED) is a promisingtechnique for delivering drugs that do not readily cross

the blood brain barrier (BBB) to neural tissue. In thisprocedure, a solution containing the drug is infusedinto the brain through a small needle or catheterinserted into the brain parenchyma. Transport of theinfusate in the brain interstitium is driven by pressuregradients rather than concentration gradients (as is thecase for di!usive delivery from an implant or otherreservoir). Therefore, infused drugs are able to pene-trate farther into the tissue.3 Since CED is often used todeliver therapeutics that are too large to cross the BBB,such as nanoparticles, liposomes, proteins, and viralvectors, it is essential to fully understand how theselarge constructs are transported through tissue. This isimportant in planning CED protocols to minimizeleakage of therapeutics away from diseased areas intohealthy tissue. Unanticipated outflow can cause seriousside effects and diminish the efficacy of the therapy, andhas been one of the major drawbacks associated withCED in clinical trials.16 The difficulty in controlling theinfusate distribution is due to the inhomogeneous nat-ure of brain tissue, which contains regions of varyingpermeability such as white and gray matter, ventricles,and necrotic regions within diseased tissue.

Recent CED studies of large therapeutics suggestthat perivascular spaces may also influence the distri-bution of infusates by providing high permeabilitypaths for fluid to move through the brain. Perivascularspaces surrounding arterioles that plunge into thecortex are extensions of the sub-pial spaces.12 For manyyears, the role of perivascular spaces has been studied inthe bulk flow of cerebrospinal fluid (CSF) through thebrain and in solute transport through the central ner-vous system.10,13,21,23,31,33 Some data suggest that per-ivascular spaces may become more prominent when

Address correspondence to Conor P. Foley, Citigroup Biomedi-cal Imaging Center, Radiology Department, Weill Cornell MedicalCollege, 516 E 72nd Street, New York, NY 10021, USA. Electronicmail: cof2003@med.cornell.edu

Annals of Biomedical Engineering (! 2011)DOI: 10.1007/s10439-011-0440-0

! 2011 Biomedical Engineering Society

Author's personal copy

endogenous particles such as amyloid-b aggregates orother particles are present.22,34

Cserr andOstrach5 examined how a bolus infusion ofblue dye-labeled dextran in the striatum of a rat dis-persed through the brain over 24 h. They found that thedye spread over a greater distance than would be pos-sible through pure diffusion, and noted that the tracertravelled away from the infusion site along perivascularspaces. Furthermore, they demonstrated that whenedema around the insertion site was eliminated (eitherby administering an anti-inflammatory or by waiting7 days after inserting the cannula before infusion), thatthe dye remained localized around the insertion site andwas not widely distributed throughout the brain. Thissuggests that perivascular spaces act as conduits forelimination of excess fluid from the brain.

Carare et al.4 found that soluble tracers such asfluorescently labeled dextran drained along perivascu-lar spaces and out of the brain. In contrast, Zhanget al.35 found that larger insoluble particulate tracers,such as amyloid-b aggregates and fluorescent micro-spheres, were trapped in the perivascular spaces. Thisaggregation may impede bulk fluid drainage at suffi-ciently high deposition levels.

Studies examining CED of liposomes and viralvectors have observed evidence of perivascular trans-port in rodents and primates.6,11,15,17,24 Mamot andcoworkers17 co-infused liposomes and mannitol intorodent brains and intracranial and flank implantedxenograft tumors. During post-mortem histologicalexamination they observed that the liposomes weredistributed along the blood vessels of the tumors.Krauze et al.15 infused gadolinium- and rhodamine-loaded liposomes into the putamen of non-humanprimates while performing MRI and magnetic reso-nance angiography (MRA). Co-localizing images fromthe MRI and MRA studies showed that the liposomeswere travelling along large blood vessels in the brain,and histological examination showed that the lipo-somes were localized in the perivascular space of thesevessels. Hadaczek and colleagues infused viral vectors,fluorescent liposomes, and BSA into the rat striatumwhile controlling the heart rate and blood pressure ofthe animal.11 They saw evidence of perivasculartransport along vessels, and observed that the finaldistribution volumes of the infusates were directlycorrelated with heart rate. They suggested that thepassage of infusates through the brain may be aided by‘‘perivascular pumping’’ through the perivascularspaces driven by pulsations of the vessel walls.

Wang and Olbricht have recently developed a modelof fluid flow in the perivascular spaces that investigatesthe possible e!ects of ‘‘perivascular pumping’’ ontransport in the perivascular spaces.30 They found thatpulsatile motion of the vessel wall could significantly

contribute to flow in the perivascular spaces even whenthere is a pressure gradient due to an ongoing CEDinfusion. They noted that the perivascular pumpingeffect became even more important as the distance ofthe perivascular space from the outlet of the CEDinfusion increased.

Although these studies demonstrate that large ther-apeutics tend to accumulate in the perivascular space, itis di"cult to garner knowledge about the dynamics oftheir motion through tissue. Studies using MRI cantrack the distribution of paramagnetically labeledinfusates in real time, but they lack the spatial resolu-tion required for examining dynamic behavior of all butthe largest blood vessels of the brain. In this study weused two-photon excited fluorescence (2PEF) micros-copy to examine how nanoparticles are transportedthrough the rat cortex during CED. This techniqueallowed us to monitor the nanoparticle distribution inreal time, and with micrometer-scale resolution.

2PEF microscopy permits fluorescence imaging withintrinsic optical sectioning deep inside scatteringspecimens with di!raction-limited resolution.8 Briefly,a femtosecond laser pulse is tightly focused inside aspecimen that has been labeled with a fluorescentmolecule that does not linearly absorb at the wave-length of the femtosecond laser. At the laser focus, thelaser intensity can become high enough to induce two-photon excitation of the fluorescent molecule. Becausethe excitation is nonlinear, the resulting fluorescence isonly produced in the focal volume where the laserintensity is high. The fluorescence intensity is thenrecorded as the position of the laser focus is scannedthroughout the specimen, forming a three-dimensionalmap of the distribution of the fluorescent label. Inaddition, because photoexcitation occurs only at thelaser focus, photobleaching of fluorescent dyes andphotodamage to the sample are reduced significantlycompared with linear imaging techniques.26

2PEF microscopy is well suited to in vivo imaging,especially deep inside highly scattering specimens. Inwidefield or confocal fluorescence microscopy, thefluorescence must be imaged to a camera or to a pin-hole, respectively. Scattering of the fluorescence lightleads to an unwanted background in widefieldmicroscopy and to decreased signal strength in con-focal microscopy. In 2PEF microscopy, however,because all the fluorescence originates from the focalvolume, it need only be detected in order to contributeto the signal, not imaged to a camera or pinhole. Thus,fluorescence that is scattered on the way to the detectorstill contributes to image formation, and does notproduce any unwanted background. This tolerance toscattering of the fluorescence allows imaging deep intoscattering samples. The imaging depth is ultimatelylimited by scattering of the femtosecond laser beam.

FOLEY et al.

Author's personal copy

In practice, one can image 500 lm or more beneath thesurface of typical tissues (brain, skin, kidney, etc.)without loss of image resolution.7,8,14,28

Further, the high spatial and temporal resolutionthat can be obtained with 2PEF microscopy makes itideal for examining dynamics of processes inside livingtissues. This technique has been used to measure andmap changes in blood flow within cortical vesselscaused by small laser-induced strokes.25 The use ofcalcium sensitive dyes and fast laser rastering tech-niques can elucidate the dynamics of activity in neu-ronal and astrocytic networks.9 Multi-photon imaginghas also been used to track cell migration and angio-genesis in rodent models of gliobastoma mulitforme.32

In this study, we infused fluorescent nanoparticlesinto the cortex of rats using microfluidic deliverydevices. During the infusions, we captured 2PEF imagesat a fixed location in the brain at a known distance fromthe tip of the microfluidic probe. By noting wherenanoparticles first appeared in the imaging plane, wecould determine whether the particles were transportedprimarily through the perivascular spaces or throughthe extracellular space (ECS). Furthermore, becauseboth the distance traveled by the nanoparticles and theelapsed time were measured, we were able to determinethe primary transport mechanism by comparing theelapsed time with the characteristic times for convectiveand di!usive transport. We found that the nanoparti-cles were transported through the perivascular spacesmuch more readily than through the bulk ECS of thebrain. These results show that delivery of large (50–150 nm diameter) therapeutic constructs to neural tissueusing CED requires careful consideration of the vas-culature in the target area to minimize the escape oftherapeutics from the disease a#icted regions and tomaximize the e!ectiveness of the infused drug.

MATERIALS AND METHODS

Nanoparticle Preparation and Characterization

Carboxylate-modified fluorescent nanoparticlestock solutions (24 and 100 nm FluoSpheres, Invitro-gen, Carlsbad CA) were sonicated for 15 min beforebeing diluted to 0.02 wt% solids in a solution of1% bovine serum albumen (BSA, Sigma-Aldrich,St. Louis, MO) in phosphate-bu!ered saline (approx-imately 1013 particles/mL and 1011 particles/mL for 24and 100 nm particles, respectively). The nanoparticlesolutions were then gently rotated using a laboratoryrotisserie at room temperature for 4 h. This treatmentreduced the surface charge of the nanoparticles andthereby reduced non-specific binding of the particles tothe extracellular matrix (ECM).19

The size and surface charge of the nanoparticles weredetermined before and after BSA incubation viadynamic light scattering and laser Doppler measure-ments of particle mobility under electrophoresis(Zetasizer Nano, Malvern Instruments Ltd., UK),respectively. The diameters of the particles determinedfrom light scattering as received from the manufacturerwere 44.5 ± 1.6 nm for the small particles (24 nmnominal diameter) and 134.8 ± 0.6 nm for the largeparticles (100 nm nominal diameter). These differencesbetween the measured and nominal diameters may bedue to slight aggregation of the sample combined withthe fact that dynamic light scattering measurements canbe skewed by the presence of a few large particles. Afterincubation with BSA, the measured diameter of theparticles increased to 62.1 ± 3.9 and 152.3 ± 2.7 nmfor the small and large nanoparticles, respectively. Theaverage increase in diameter of the particles is 17.6 nm,which is in good agreement with the hydrodynamic sizeof BSA (7.2 nm2). The zeta potentials of the nanopar-ticles before BSA coating were 236.8 ± 2.8 and238.6 ± 3.6 mV for the small particles and large par-ticles, respectively. After BSA coating the zeta poten-tials were reduced to 210.8 ± 3.4 mV for the small,and 211.4 ± 2.5 mV for the large particles. As antici-pated, the surface charge was substantially decreasedby the presence of BSA.

Microfluidic Device Preparation

Silicon-based microfluidic devices were fabricated aspreviously described.18 The devices had a 5-mm-longinsertable shank with a 100 9 100 lm cross section.The microfluidic probes were attached to 150 mm longborosilicate micropipettes (1 mm OD, 0.58 mm ID)(World Precision Instruments Inc., Sarasotam, FL)with two-part epoxy (Epoxy 907, Miller-Stephenson,Danbury, CT). The devices were backfilled with thenanoparticle solution under vacuum before the glasscapillary was filled via a syringe with a customMicroFil tip (World Precision Instruments Inc.,Sarasota, FL). The micropipette served as the fluidicreservoir for the microfluidic device. The proximal endof the glass micropipette was connected to a pro-grammable pressure injector (PM8000, World Preci-sion Instruments Inc.) using a micro-electrode holder.To infuse fluid using the microfluidic device, a constantpressure was applied to the fluidic reservoir in themicropipette via the pressure injector.

In Vivo Nanoparticle Infusions

Five male Sprague–Dawley rats (weighing 283–370 g) were anesthetized via intra-peritoneal injectionof 30% wt/vol urethane in deionized water (urethane

Real-Time Imaging of Perivascular Transport of Nanoparticles

Author's personal copy

dose of 150 mg/100 g body weight). Animals were thensecured in a stereotaxic frame and an incision wasmade in the skin along the dorsal midline to expose theskull. A dental drill was used to open a large craniot-omy (~1-cm diameter), exposing the brain from nearthe midline to the zygomatic arch. The dura wasremoved using a fine hook. The vasculature waslabeled via tail vein injection of either fluoresceinisothiocyanate (FITC) conjugated to 2-MD dextran ortetramethylrhodamine conjugated to 155 kDa dextran(Sigma-Aldrich, St. Louis, MO). The tail vein injectionfluorescently labels blood serum, allowing blood ves-sels of the cortex to be visualized under 2PEF. Whenred fluorescent nanoparticles were used, the bloodvessels were labeled with green FITC-dextran; whengreen fluorescent nanoparticles were used, the vascu-lature was labeled with red tetramethylrhodamine-dextran.

Animals were then transferred to the stage of the2PEF microscope, and microfluidic probes wereinserted 1 mm into the cortex at a downward angle of15" using a micromanipulator. The tissue was allowedto equilibrate for at least 2 min before infusions began.During this time, imaging stacks were collected in thedorsoventral direction to map the vasculature and todetermine a suitable imaging location for monitoringthe nanoparticle distribution during the infusion. Theselected imaging location was the level directly abovethe probe outlet that had the greatest number of ver-tically oriented blood vessels (i.e., penetrating arteri-oles and ascending venules) that intersected theimaging frame. Arterioles and venules were easily dis-tinguished on the cortical surface by their morphologyand flow direction. To categorize subsurface vessels,the vessels were traced to readily identifiable surfacearterioles and venules. A schematic of the experimentalset-up is shown in Fig. 1. Infusions were started at adriving pressure of 0.5 psi. The infusion pressure wasincreased at a rate of 0.1 psi/30 s to a final pressure of1 psi. In bench-top test infusions in agarose phantomsthis resulted in a flow rate of about 0.1 lL/min, whichwas sufficiently low to prevent backflow around thedevice. However, during imaging of in vivo infusions,direct measurement of the actual flow rate was pre-cluded by the position of the microfluidic device withinthe stage of the multiphoton microscope. 2PEF imag-ing frames were captured at 3.4 frames/s throughoutthe infusions and examined for the location of the firstappearance of nanoparticle fluorescence (in perivas-cular spaces or in ECS). After an infusion was com-plete, the imaging plane was scanned through the tissuein the dorsoventral direction to sample the finalnanoparticle distribution (up to a maximum depth ofapproximately 700 lm below the brain surface). Twoanimals received a second infusion of nanoparticles

within the same craniotomy; the second infusion siteswere chosen so that no residual fluorescence could beseen from the first infusion (>5 mm distance betweensites).

2PEF images were taken on a two-channel micro-scope of local design using 100 fs duration pulses from a76 MHz repetition rate Ti:sapphire oscillator (Mira-HP, Coherent, Santa Clara, CA) pumped by a contin-uous wave laser (Verdi-V18, Coherent, Santa Clara,CA). Excitation wavelengths were centered at 810 and850 nm for yellow–green and red fluorescent nanopar-ticles, respectively. Detection filters were 525/70 nm(center wavelength/spectral width) and 615/70 nm. Lowmagnification images were taken with a 0.28 numericalaperture (NA), 49 air objective (Olympus). High mag-nification images used a 0.95 NA, 209 water immersionobjective (Olympus) with a 2 mm working distance.

All animal procedures were carried out in accor-dance with the Cornell University Institutional AnimalCare and Use Committee guidelines and regulations.

Image Analysis

2PEF imaging files were exported to ImageJ1 asmulti-image tagged image file format (TIFF) stacksfor analysis. Each channel of the 2PEF image (one forvasculature, one for nanoparticles) was adjusted forbrightness and contrast and combined into a two-colorimaging stack for viewing and export.

Where applicable, spot areas were measured by firstimporting the single-channel 2PEF file containing thenanoparticle fluorescence data to ImageJ as a TIFF.This file was adjusted for brightness and contrast, andthen converted to binary where the nanoparticle fluo-rescence was bright and the background was dark.The threshold for this operation was determined bythe ImageJ ‘‘getAutoThreshold’’ function using the

FIGURE 1. Experimental set-up for real-time two-photonexcited fluorescence imaging of perivascular transport ofnanoparticles.

FOLEY et al.

Author's personal copy

IsoData algorithm. Next, noise was removed from thebinary image by applying a 2 9 2 pixel median filter.The total area of the white pixels in each frame of theimaging stack could then be measured to assess thechange in spot area over time.

RESULTS

Nanoparticle Transport in the Rat Cortex

We used 2PEF microscopy to examine in real timehow BSA-coated polystyrene fluorescent nanoparticlesare transported through the cortex of live, anesthetizedrats during CED. Microfluidic probes were inserted1 mm into the cortex of rats at a downward angle ofapproximately 15", and nanoparticle solutions wereinfused using the ramped pressure profile describedabove (0.5–1 psi over the first 150 s of the infusion).Imaging frames were captured during infusions at afixed level within the cortex, directly above the probeoutlet (Fig. 1). We found that nanoparticles of bothsizes traveled preferentially along perivascular spacesrather than through the ECS. Table 1 summarizesexperimental conditions and results for each infusion,giving the particle size, the distance from the micro-fluidic probe outlet to the imaging plane, and the timetaken for nanoparticles to reach the imaging plane.

24 nm Nanoparticle Transport

Infusions 1 through 5 in Table 1 involve 24 nmBSA-coated nanoparticles. For infusion 1, we chose animaging plane 94 lm above the outlet of the micro-fluidic device. Nanoparticles were seen to transportthrough the ECS, and no perivascular transport wasobserved in the imaging plane during the infusion.Post-infusion imaging up to 5 min after the end of theinfusion showed that nanoparticles had localized inperivascular spaces in a more dorsal plane than theimaging plane used during the infusion.

In cases 2 through 5, nanoparticles were observed inthe imaging plane shortly after the infusion started. In

these instances, the nanoparticles first appeared in theperivascular spaces of blood vessels that intersected theimaging plane. The particles later filled the ECSbetween the vessels. Figure 2 shows a sequence ofimages recorded at times of 0, 30, 90, and 150 s afterthe start of the infusion. The first image (Fig. 2a)shows the imaging plane at the start of the infusion.Fluorescently labeled blood vessels (green) that inter-sect the imaging plane can be seen, but there are nonanoparticles present because they are released fromthe probe that is 243 lm below the imaging plane. Inthe second image (Fig. 2b), fluorescently labelednanoparticles (red) can be seen in the perivascularspaces of a few blood vessels in the lower left part ofthe image. In the third panel (Fig. 2c), more of theblood vessels are surrounded by nanoparticles. Othernanoparticles have started to fill the space betweenvessels in Figs. 2c and 2d. The nanoparticles did notspread into the tissue in a direction perpendicular tothe penetrating vessels in appreciable amounts on thetime scale of our experiments. The fluorescence in theperivascular space was primarily constrained aroundarterioles, with a small amount of fluorescenceextending a short distance (~40–50 lm) along capil-laries that branch from penetrating arterioles. This canbe seen in detail in Fig. 3. This distribution of peri-vascular fluorescence could be expected, because weselected our imaging plane to have the greatest possiblenumber of vertically penetrating cortical vessels.Researchers have demonstrated that particulatesinjected into sub-pial spaces and cortex are likely to betrapped in the perivascular spaces of arterioles.4 Theimaging data we collected represents a small fraction ofthe overall nanoparticle distribution in the tissue.Therefore, it is difficult to estimate the fraction ofnanoparticles that were transported in perivascularspaces.

In the infusion that took 200 s (infusion 3), post-infusion analysis showed that the majority of infusedfluid had travelled along the insertion track of themicrofluidic device. A mechanical disturbance to the

TABLE 1. Summary of the results of the nanoparticle infusions, giving the size of the particle infused, the distance from the probetip to the imaging plane, and the time taken for the particles to appear in the imaging frame.

InfusionNominal particle

size (nm)Distance to imaging

plane (lm)

Time to arrival at imaging plane(s)

FigurePervivascular ECS

1 24 94 No NP seen at imaging plane –2 24 243 13 50 Fig. 23 24 290 200 N/A –4 24 220 12 35 Fig. 35 24 360 30 135 Fig. 46 100 300 N/A 110 –7 100 280 No NP seen at imaging plane Fig. 5

Real-Time Imaging of Perivascular Transport of Nanoparticles

Author's personal copy

microfluidic probe resulted in tearing of the tissuearound the device tip. Most of the infused fluid wastransported along the high permeability device track,reducing the driving pressure to force fluid through theECS to the perivascular space. The perivasculartransport observed at the fixed imaging plane duringthe infusion was along a vessel that passed close to themicrofluidic probe in a more ventral plane, and therebyintersected the backflowing nanoparticles in the inser-tion track.

For the other infusions (2, 4, and 5), the time forparticles to arrive at the imaging plane was between13 and 30 s. Fluorescence first appeared in the

perivascular spaces, before gradually filling in the ECS(arrival time of 35–135 s; Table 1).

In infusion 4, we observed the motion of nanopar-ticles along an arteriole (approximately 30 lm indiameter) that lay in the imaging plane at a distance of360 lm above the tip of the probe (Fig. 4). The mea-sured width of the perivascular space around this vesselafter it filled with nanoparticles was 8–10 lm.

100 nm Nanoparticle Transport

Infusions 6 and 7 in Table 1 involve 100 nmBSA-coated nanoparticles. In infusion 6 we saw

FIGURE 2. Time course showing transport of fluorescent nanoparticles (red, 24 nm nominal diameter) through perivascularspaces during CED. Fluorescently labeled blood vessels are shown in green. Images were captured 243 lm above the outlet of themicrofluidic device. (a) Infusion time 5 0 s; (b) 30 s; (c) 90 s; (d) 150 s. Note appearance of nanoparticles around vessels in panel(b), and gradual filling in of background ECS in (c) and (d). The dark band across the image from top right to bottom left is due to alarge blood vessel on the surface of the brain that obscures the imaging below.

FOLEY et al.

Author's personal copy

nanoparticles reach the imaging plane (a distance of300 lm) 110 s after the infusion started. In infusion 7,nanoparticles were not observed at the fixed imagingplane during the infusion. However, post-infusionanalysis revealed that the infused particles followed theperivascular space of a vessel that passed close to theoutlet of the microfluidic device, but did not intersectthe fixed imaging frame. Figure 5 shows a series ofimaging planes in the dorso-ventral direction from thisinfusion. The figure shows 100 nm particles in theperivascular space of an arteriole that stretches over adistance of more than 150 lm. Post-infusion imagingshowed that in both infusions the nanoparticles did notmove extensively through the ECS, but, instead, wereconfined in the perivascular space of vessels (Fig. 5).

DISCUSSION

24 nm Nanoparticle Transport

During infusion 1, we observed a growing spot offluorescence caused by convection of the nanoparticlesthrough the ECS. This fluorescent area represented aplane section through an infusion volume (VINF) thatincreased with time. If we assume that the infusion isisotropic into a porous medium without perivasculartransport, we can use data from the experiment to

obtain an estimate of the flow rate of the fluorescentnanoparticles through the ECS. This procedure prob-ably overestimates the actual flow rate through theECS because it ignores some transport through theperivascular space. Nevertheless, we can use this esti-mated flow rate to compute the expected arrival timesat the imaging plane in our experiments in the absenceof perivascular transport and compare the expectedvalues with our measurements to gain insight into thesignificance of perivascular spaces on nanoparticletransport.

Under the assumption of an isotropic infusion intoa porous medium from a point source, the infusionvolume is spherical, and the plane section is a circle(see Fig. 6a). The area of this plane section is given by:

APLANE ! p R2INF " d2

! "#1$

where APLANE is the cross-sectional area of theobserved spot of fluorescence, RINF is the radius ofthe spherical infusion volume, and d is the distance ofthe plane section from the center of the infusion vol-ume, which is 94 lm in this case.

Rearranging Eq. (1) yields:

RINF !APLANE

p% d2

# $1=2

#2$

The infusion volume at time, t, is given by:

VINF !4

3pR3

INF !Qt

/#3$

where Q is the volumetric infusion rate, and / is theporosity of brain tissue (~0.220). Combining Eqs. (2)and (3) gives:

3Qt

4p/! APLANE

p% d2

# $3=2

#4$

Plotting the right hand side of Eq. (4) (i.e., R3INF) as

a function of time yields a straight line of slope, m,where:

m ! 3Q

4p/#5$

Using APLANE as measured from our imaging stackand d = 0.094 mm (Fig. 6b), we can estimate the flowrate of nanoparticles through the ECS for this infu-sion to be 0.008 lL/min. This approximation is sub-ject to the following assumptions: (1) effects ofperivascular transport are neglected; (2) there is min-imal diffusion of the nanoparticles on the timescale ofthe imaging experiments, which is reasonable giventhat the nanoparticle diameter is close to the reportedpore size of the ECM29; and (3) the concentration ofnanoparticles is uniform in the measured spot, which

FIGURE 3. Image showing 24 nm nanoparticles (red) inperivascular spaces around a penetrating arteriole andbranching capillary (blood vessels shown in green). BSA-coated red fluorescent nanoparticles were infused into the ratcortex, and the vasculature was labeled with FITC-dextran.The image shows the distribution of fluorescent nanoparticlearound a branching capillary (~10 lm diameter). Nanoparti-cles extend for ~50 lm along the capillary after branchingfrom the arteriole.

Real-Time Imaging of Perivascular Transport of Nanoparticles

Author's personal copy

also is reasonable given that the imaging plane is 2 lmthick.

Studies of particle di!usion in the brain indicatethe pore size of the ECM is between 38 and 64 nm,29

which is comparable to the particle size in thisexperiment. Interactions between the suspended par-ticles and constituents of the ECM can hinder themotion of particles, resulting in a longer thanexpected convective time. Previous studies by ourgroup involving CED infusions of identical 24 nmnanoparticles into the rat striatum examined thepenetration of infused nanoparticles into gray matterand compared it with the penetration of infusedfluid. The results showed that the nanoparticles move

through the tissue much more slowly than the infusedfluid, which indicates that particle motion is stronglyhindered.19

If we again consider the infusion to be an isotropicinfusion into a porous medium from a point source,then the characteristic time tc for the convective frontof infused fluid to reach a radial distance r from thepoint source is:

tc !4p/3Q

r3 #6$

Using this equation with the estimated flow rate of0.008 lL/min, we find that the characteristic times fora convective front of infused nanoparticles to reach the

FIGURE 4. Time course showing transport of fluorescent nanoparticles (red, 24 nm nominal diameter) along a vessel in theimaging plane. The vasculature is labeled with FITC-dextran. Times represent duration of the infusion (a) 5 14.5 s, (b) 5 34.8 s,(c) 5 41.0 s, and (d) 5 65.5 s. Images were captured at a plane 360 lm above the outlet of the microfluidic probe, 240 lm below thesurface of the brain.

FOLEY et al.

Author's personal copy

imaging planes in our experiments (distances of 220–360 lm) are between 67 and 292 s.

The characteristic time td for a particle to diffuse adistance r through brain tissue can be determined from:

td !r2

D&#7$

where D* is the effective diffusion coefficient of theparticle in the tissue given by:

D& ! D

s2#8$

where D is the diffusion coefficient of the particle in afree medium and s is the tortuosity of the ECS of the

brain (s ~ 227,29). The diffusion coefficient of a particlein a free medium (e.g., water) can be estimated fromthe Stokes–Einstein equation:

D ! kT

3pdHg#9$

where k is the Boltzmann constant (1.38065 910223 m2 kg22 K21), T is the temperature in Kelvin(310 K), dH is the hydrodynamic diameter of the par-ticle (24 nm for the largest possible value of D), and gis the viscosity of the free medium at the temperature T(6.915 9 1024 Pa s29). Using these equations, andassuming that the nanoparticles are able to diffusethrough the pores of the ECM, we find that the

FIGURE 5. Sections from a post-infusion imaging stack in the dorsoventral direction, showing red fluorescent 100 nm nano-particles constrained in the perivascular space (vessels shown in green). Frame (a) is an optical section 50 lm below the brainsurface; (b) 100 lm; (c) 150 lm; (d) 200 lm. Images show that the nanoparticles are distributed in the perivascular space of thevessel over a distance of several hundred micrometers. The outlet of the microfluidic device was 550 lm below the brain surface.

Real-Time Imaging of Perivascular Transport of Nanoparticles

Author's personal copy

characteristic time required for a particle to reach ourimaging plane from the outlet of the probe, a distanceof 220–360 lm, by pure diffusion is between 90 and241 min.

In our experiments, the time taken for fluorescenceto appear at the imaging plane ranged from 13 to 30 s(infusions 2, 4, and 5; Table 1), which is much shorterthan the characteristic time for hindered convec-tive transport (67–292 s) or diffusive transport(90–241 min). Note that infusion 3 was not includeddue to the fact that a mechanical disturbance causedthe infusate to backflow around the shank of thedevice. This suggests that convective transport throughthe perivascular spaces allows these particles to movethrough brain tissue much more rapidly than by con-vection or diffusion through the ECS.

In infusion 4 we observed nanoparticles movingthrough the perivascular spaces of an arteriole that layin the imaging plane (Fig. 4; Movie 2 in Supplemen-tary material). This in-plane view of the vesselunequivocally demonstrates that nanoparticles areconstrained in, and rapidly travelling along, the peri-vascular spaces of the vessels.

100 nm Nanoparticle Transport

In infusion 6, 100 nm particles reached the imagingplane (a distance of 300 lm) 110 s after the infusionstarted, which was approximately 4 times longer thanrequired for 24 nm particles. The long time taken forthe nanoparticles to appear at the imaging plane islikely due to greatly hindered transport through theECS, since their diameter is about twice the estimatedaverage pore size of the ECM.29 However, once thenanoparticles reach the perivascular space, there islittle structure present to impede their motion.21 Thissuggests that perivascular spaces are more importanttransport routes for particles larger than the pore sizeof the ECS than for smaller particles that undergo lesshindered transport in the ECS.

It seems clear from these limited data that thespread of the 100 nm particles in the ECS is much lessthan the corresponding spread of the 24 nm particles.This suggests that the fraction of particles that movedthrough the perivascular space rather than through theECS was larger for the 100 nm particles than for the24 nm particles, which could explain the relativelyinhomogeneous distribution observed for the largerparticles.

Implications for CED-Based Therapies

These results suggest that it is essential to considere!ects of perivascular transport when planning CEDprotocols. E!ects of perivascular transport may bebeneficial in some cases, but detrimental in others.2PEF studies of cancer cells have shown that migratingglioma cells travel farther and faster in perivascularspaces than they do through the ECS, and that thesecells promote angiogenesis from the blood vessels theytrack.32 These are the cells that are most likely to beleft behind after tumor resection, leading to tumorrecurrence. One could envision a post-resection CEDtreatment that exploits the hydrodynamic properties ofnanoparticles to preferentially track and target infil-trating glioma cells in the perivascular spaces of theperitumoral tissue. A thorough understanding of thetarget tissue vasculature and the in vivo transportcharacteristics of the therapy would be required toexploit this strategy.

Conversely, there may be situations where perivas-cular transport of infused therapeutics is undesirable.If a CED therapy involves delivering constructs thatundergo hindered transport through the ECM to atarget where perivascular spaces are present, thoseperivascular spaces may provide a high permeabilitypath of undesired egress into healthy tissue. In such acase, it may be of use to increase the hydraulic per-meability of the bulk ECS to reduce the likelihood that

(a)

(b)

FIGURE 6. (a) Schematic showing geometry used in deter-mining the flow rate of nanoparticles. The blue sphere repre-sents the volume of infusion, and the red plane represents thearea measured in our imaging stacks. (b) Plot of R3

INF (seeEq. 2) as a function of time, for infusion of 24 nm nanoparti-cles (infusion 1). The slope of 0.0096 lL/min allows us todetermine the nanoparticle flow rate to be 0.008 lL/min.

FOLEY et al.

Author's personal copy

the particles will travel along the blood vessels tohealthy tissue. The hydraulic permeability can bemodified through degrading or dilating the ECM asdemonstrated by Neeves and coworkers.19 An alter-native way to decrease the fraction of the therapeuticthat escapes the target area through perivascularspaces would be to make the constructs smaller,thereby increasing the apparent permeability of theECM seen by the particles.

The use of 2PEF to observe the dynamics of peri-vascular transport provides high resolution andquantitative information that cannot be obtained withother imaging techniques. However, there are limita-tions to the use of 2PEF that should be mentioned.Even for the relatively low infusion rates and lowinfusion volumes used in these experiments, it was notpossible to map the entire infusate distribution in theparenchyma. Since the infusate distribution is roughlysymmetrical about the outlet of the device, some of theinfusate moves too deep into the tissue for us to image.Further, the imaging plane we use covers only a smallcross section (up to ~2 mm2) of the injected volume.Without rapidly moving the imaging plane in all threedimensions, it is impossible to image the infusate dis-tribution outside of the imaging plane. If the front ofinfused fluid reaches a fluid sink, such as the surface ofthe brain or a ventricle, it would not be captured in the2PEF image. Despite these limitations, the technique isa still a powerful method to assess particle and infusatedynamics at the leading edge of an infusion volume,since no other in vivo imaging technique providescomparable temporal and spatial resolution.

Conclusions

We used 2PEF microscopy to study the transport offluorescent nanoparticles through the cortex of ratsduring CED via microfluidic probes. These experi-ments represent, to our knowledge, the first use of2PEF microscopy for real time in vivo tracking ofinfusates during CED, and show that 2PEF micros-copy is a powerful tool for identifying transportcharacteristics of fluorescently labeled infusatesthrough tissue. Furthermore, we present a viabletechnique for measuring the in vivo parenchymal flowrate of infused compounds. We examined the distri-butions of two different sizes of nanoparticles, andobserved that perivascular spaces in the area of themicrofluidic device outlet can greatly affect the trans-port of the particles by providing high permeabilityconduits for fluid transport.

Furthermore, we noted that perivascular transportis dependent on the size of the infused nanoparticle.For smaller particles that can readily pass through theECM, the fraction of the infused particles that move

through the perivascular spaces rather than throughthe ECS is smaller than for large particles that undergosignificantly hindered transport in the ECM.

From these results, it is clear that a thoroughunderstanding of the vasculature of the target tissueand the transport properties of the therapy beingdelivered are of critical importance in designing e!ec-tive CED protocols.

ELECTRONIC SUPPLEMENTARY MATERIAL

The online version of this article (doi:10.1007/s10439-011-0440-0) contains supplementary material,which is available to authorized users.

ACKNOWLEDGMENTS

This work was supported by the National Institutesof Health (Grant NS-045236 to WLO), the EllisonMedical Foundation (Grant AG-NS-0330-06 to CBS),and the American Heart Association (Grant 0735644Tto CBS). This work was performed in part at theCornell NanoScale Facility, a member of the NationalNanotechnology Infrastructure Network, which issupported by the National Science Foundation (GrantECS-0335765). Also, this work made use of STCshared experimental facilities supported by theNational Science Foundation under Agreement No.ECS-9876771.

REFERENCES

1Abramoff, M. D., P. J. Magelhaes, and S. J. Ram. Imageprocessing with ImageJ. Biophoton. Int. 11:36–42, 2004.2Bezemer, J. M., D. W. Grijpma, P. J. Dijkstra, C. A. vanBlitterswijk, and J. Feijen. A controlled release system forproteins based on poly(ether ester) block-copolymers:polymer network characterization. J. Control Release62:393–405, 1999.3Bobo, R. H., D. W. Laske, A. Akbasak, P. F. Morrison,R. L. Dedrick, and E. H. Oldfield. Convection-enhanceddelivery of macromolecules in the brain. Proc. Natl. Acad.Sci. USA 91:2076–2080, 1994.4Carare, R. O., M. Bernardes-Silva, T. A. Newman, A. M.Page, J. A. R. Nicoll, V. H. Perry, and R. O. Weller. Sol-utes, but not cells, drain from the brain parenchyma alongbasement membranes of capillaries and arteries: signifi-cance for cerebral amyloid angiopathy and neuroimmu-nology. Neuropathol. Appl. Neurobiol. 34:131–144, 2008.5Cserr, H. F., and L. H. Ostrach. Bulk flow of interstitialfluid after intracranial injection of blue dextran 2000. Exp.Neurol. 45:50–60, 1974.6Cunningham, J., P. Pivirotto, J. Bringas, B. Suzuki,S. Vijay, L. Sanftner, M. Kitamura, C. Chan, and K. S.Bankiewicz. Biodistribution of adeno-associated virus

Real-Time Imaging of Perivascular Transport of Nanoparticles

Author's personal copy

type-2 in nonhuman primates after convection-enhanceddelivery to brain. Mol. Ther. 16:1267–1275, 2008.7Denk, W., K. R. Delaney, D. Kleinfeld, B. Strowbridge,D. W. Tank, and R. Yuste. Anatomical and functionalimaging of neurons and circuits using two photon laserscanningmicroscopy. J.Neurosci.Methods 54:151–162, 1994.8Denk, W., J. H. Strickler, and W. W. Webb. Two-photonlaser scanning fluorescence microscopy. Science 248:73–76,1990.9Gobel, W., B. M. Kampa, and F. Helmchen. Imagingcellular network dynamics in three dimensions using fast3D laser scanning. Nat. Methods 4:73–79, 2007.

10Gregory, T. F., M. L. Rennels, O. R. Blaumanis, andK. Fujimoto. A method for microscopic studies of cerebralangioarchitecture and vascular-parenchymal relationships,based on the demonstration of ‘paravascular’ fluid path-ways in the mammalian central nervous system. J. Neuro-sci. Methods 14:5–14, 1985.

11Hadaczek, P., Y. Yamashita, H. Mirek, L. Tamas, M. C.Bohn, C. Noble, J. W. Park, and K. Bankiewicz. The‘‘perivascular pump’’ driven by arterial pulsation is apowerful mechanism for the distribution of therapeuticmolecules within the brain. Mol. Ther. 14:69–78, 2006.

12Hutchings, M., and R. O. Weller. Anatomical relationshipsof the pia mater to cerebral blood vessels in man. J. Neu-rosurg. 65:316–325, 1986.

13Ichimura, T., P. A. Fraser, and H. F. Cserr. Distribution ofextracellular tracers in perivascular spaces of the rat brain.Brain Res. 545:103–113, 1991.

14Kleinfeld, D., P. P. Mitra, F. Helmchen, and W. Denk.Fluctuations and stimulus-induced changes in blood flowobserved in individual capillaries in layers 2 through 4 ofrat neocortex. Proc. Natl. Acad. Sci. USA 95:15741–15746,1998.

15Krauze, M. T., R. Saito, C. Noble, J. Bringas, J. Forsayeth,T. R. McKnight, J. Park, and K. S. Bankiewicz. Effects ofthe perivascular space on convection-enhanced delivery ofliposomes in primate putamen. Exp. Neurol. 196:104–111,2005.

16Lidar, Z., Y. Mardor, T. Jonas, R. Pfeffer, M. Faibel,D. Nass, M. Hadani, and Z. Ram. Convection-enhanceddelivery of paclitaxel for the treatment of recurrent malig-nant glioma: a phase I/II clinical study. J. Neurosurg.100:472–479, 2004.

17Mamot, C., J. B. Nguyen, M. Pourdehnad, P. Hadaczek,R. Saito, J. R. Bringas, D. C. Drummond, K. Hong, D. B.Kirpotin, T. McKnight, M. S. Berger, J. W. Park, and K. S.Bankiewicz. Extensive distribution of liposomes in rodentbrains and brain tumors following convection-enhanceddelivery. J. Neurooncol. 68:1–9, 2004.

18Neeves, K. B., C. T. Lo, C. P. Foley, W. M. Saltzman, andW. L. Olbricht. Fabrication and characterization ofmicrofluidic probes for convection enhanced drug delivery.J. Control Release 111:252–262, 2006.

19Neeves, K. B., A. J. Sawyer, C. P. Foley, W. M. Saltzman,and W. L. Olbricht. Dilation and degradation of the brainextracellular matrix enhances penetration of infused poly-mer nanoparticles. Brain Res. 1180:121–132, 2007.

20Nicholson, C. Diffusion and related transport mechanismsin brain tissue. Rep. Prog. Phys. 64:815–884, 2001.

21Patek, P. The perivascular spaces of the mammalian brain.Anat. Rec. 88:1–24, 1944.

22Preston, S. D., P. V. Steart, A. Wilkinson, J. A. R. Nicoll,and R. O. Weller. Capillary and arterial cerebral amyloidangiopathy in Alzheimer’s disease: defining the perivascularroute for the elimination of amyloid beta from the humanbrain. Neuropathol. Appl. Neurobiol. 29:106–117, 2003.

23Rennels, M. L., T. F. Gregory, O. R. Blaumanis,K. Fujimoto, and P. A. Grady. Evidence for a ‘paravas-cular’ fluid circulation in the mammalian central nervoussystem, provided by the rapid distribution of tracer proteinthroughout the brain from the subarachnoid space. BrainRes. 326:47–63, 1985.

24Saito, R., M. T. Krauze, C. O. Noble, D. C. Drummond,D. B. Kirpotin, M. S. Berger, J. W. Park, and K. S.Bankiewicz. Convection-enhanced delivery of Ls-TPTenables an effective, continuous, low-dose chemotherapyagainst malignant glioma xenograft model. Neuro Oncol.8:205–214, 2006.

25Schaffer, C. B., B. Friedman, N. Nishimura, L. F.Schroeder, P. S. Tsai, F. F. Ebner, P. D. Lyden, andD. Kleinfeld. Two-photon imaging of cortical surfacemicrovessels reveals a robust redistribution in blood flowafter vascular occlusion. PLoS Biol. 4:e22, 2006.

26Squirrell, J. M., D. L. Wokosin, J. G. White, and B. D.Bavister. Long-term two-photon fluorescence imaging ofmammalian embryos without compromising viability. Nat.Biotechnol. 17:763–767, 1999.

27Stroh, M., W. R. Zipfel, R. M. Williams, W. W. Webb, andW. M. Saltzman. Diffusion of nerve growth factor in ratstriatum as determined by multiphoton microscopy. Bio-phys. J. 85:581–588, 2003.

28Svoboda, K., W. Denk, D. Kleinfeld, and D. W. Tank. Invivo dendritic calcium dynamics in neocortical pyramidalneurons. Nature 385:161–165, 1997.

29Thorne, R. G., and C. Nicholson. In vivo diffusion analysiswith quantum dots and dextrans predicts the width of brainextracellular space. Proc. Natl. Acad. Sci. USA 103:5567–5572, 2006.

30Wang, P., and W. L. Olbricht. Fluid mechanics in theperivascular space. J. Theor. Biol. 274:52–57, 2011.

31Weed, L. The absorption of cerebrospinal fluid into thevenous system. Am. J. Anat. 31:191–221, 1923.

32Winkler, F., Y. Kienast, M. Fuhrmann, L. V. Baumgarten,S. Burgold, G. Mitteregger, H. Kretzschmar, and J. Herms.Imaging glioma cell invasion in vivo reveals mechanisms ofdissemination and peritumoral angiogenesis. Glia 57:1306–1315, 2009.

33Woollam, D. H., and J. W. Millen. The perivascular spacesof the mammalian central nervous system and their relationto the perineuronal and subarachnoid spaces. J. Anat.89:193–200, 1955.

34Zhang, E. T., C. B. Inman, and R. O. Weller. Interrela-tionships of the pia mater and the perivascular (Virchow-Robin) spaces in the human cerebrum. J. Anat. 170:111–123, 1990.

35Zhang, E. T., H. K. Richards, S. Kida, and R. O. Weller.Directional and compartmentalised drainage of interstitialfluid and cerebrospinal fluid from the rat brain. ActaNeuropathol. 83:233–239, 1992.

FOLEY et al.

Author's personal copy